[yocto] Per image customizations

[Paul Eggleton]

On Tue, 07 Jun 2016 16:07:26 Edward Wingate wrote:
> On Tue, Jun 7, 2016 at 1:47 PM, Paul Eggleton
> <paul.eggleton at linux.intel.com> wrote:
> > So what this will give you is alternative packages to be installed instead
> > of sysvinit-inittab in your image.
> 
> I was under the (mistaken) impression that my recipe will run in
> addition to, not instead of, sysvinit-inittab (occurring after it and
> overwriting its inittab with my custom version).

The recipe will build in addition to sysvinit-inittab yes, but do_install for 
that recipe is just assembling the files ready for packaging - it doesn't 
directly install files into the image. When it comes time to assemble the 
image, all that is being done when you distil it down is that we install a 
bunch of packages.

> > Have you verified (by way of the image manifest written next to the image
> > file) that this package is actually being installed?
> 
> The manifest shows both sysvinit-inittab and my package being
> installed in the image, but sysvinit-inittab's inittab apparently
> takes precedence over mine.

Hmm, that is strange, because as they install the same file they ought to 
conflict. Perhaps as suggested elsewhere in this thread this is a bug (or at 
least undesirable behaviour) in rpm.

> > What steps have you taken to ensure that it does get installed in
> > preference to sysvinit-inittab?
>
> None, due to my mistaken impression.  After investigating your kind
> inference with help of Yocto manual, I replaced in my recipe:
> DEPENDS += " sysvinit-inittab"
> with:
> RREPLACES_${PN} = "sysvinit-inittab"
> 
> and it appears to be working now.  I've been using Yocto for a year
> now, but still feel like I'm barely scratching the surface of it.  I
> tend to get things to work by floundering about, so even though it
> does works, I would appreciate knowing if this was the most proper
> course of action.  Thanks!

I guess here you are getting into functionality that is not very well-defined, 
and you're also dealing with behaviour of the package manager rather than the 
build system itself (though I appreciate the lines are blurred from usage 
perspective). I guess it would help though if we would document a working 
procedure to do this kind of thing as it does come up from time to time.

Cheers,
Paul

-- 

Paul Eggleton
Intel Open Source Technology Centre